Google is using Anthropic’s Claude to improve Gemini AI responses, the company explains

Google uses Anthropic’s Claude AI model to evaluate the performance of its own Gemini AI, a report said, raising questions about the practice’s compliance with Anthropic’s terms and conditions.
TechCrunch claims to have viewed internal correspondence that suggests entrepreneurs working at Gemini are comparing its responses to those generated by Claude. These contractors are said to be tasked with evaluating the accuracy and quality of Gemini’s output based on criteria such as truthfulness and verbosity.
“Contractors are given up to 30 minutes per request to determine whose answer is better, Gemini’s or Claude’s,” the report said.

The difference between the answers of Gemini and Claude

TechCrunch also says contractors noticed explicit references to Claude within Google’s internal platform used to compare AI models.
In some cases, Claude’s responses appear to prioritize safety more than Gemini’s, refusing to respond to requests deemed unsafe or generating more cautious responses, the report added. In one instance, Gemini’s response was flagged as a “huge breach of security” for including “nudity and slavery”.
“Claude’s safety settings are the strictest” of the AI ​​models, an entrepreneur was quoted as saying.

What Anthropic terms say and Google’s clarification

Anthropic’s terms and conditions explicitly prohibit using Claude to “build a competing product or service” or “train competing AI models” without prior approval.
A Google DeepMind the spokesperson confirmed they were comparing model results for evaluation purposes, but denied using anthropomorphic models to train Gemini. Google is also a major investor in Anthropic.
“Of course, in accordance with standard industry practice, in some cases we compare model results as part of our evaluation process,” said Shira McNamara, a spokeswoman for Google DeepMind.
“However, any suggestion that we used anthropomorphic models to train Gemini is inaccurate,” McNamara added.